Download Location-based decision support for user groups

XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
Vol. ??, No. ?, Month?? 2009, 1–23
Location-based decision support for user groups
Martin Espetera and Martin Raubalb*
a
Institute for Geoinformatics, University of Münster, Münster, Germany; bDepartment of
Geography, University of California, Santa Barbara, USA
5
10
15
20
25
30
35
(Received 1 March 2008; final version received 27 August 2009)
The ongoing progression of mobile communication devices, in particular
with respect to positioning capabilities, has enabled the development of
a new generation of location-based services (LBS). One of the most
important aspects of current research on LBS is personalisation but it has
been treated primarily for individual users. This article presents a first step
towards personalised LBS that support user groups in their everyday
decision-making, such as deciding where and when to meet for a common
activity. The principles behind location-based group decision processes are
outlined and a formal model for supporting such processes using methods
of multi-criteria decision-making, time geography and similarity measurement is defined. This model is then implemented in the LBS prototype
mediatrix. The prototype is used for simulating a location-based decision
process aimed at finding an optimal restaurant for a user group in a city.
Keywords: location-based service; multi-criteria decision-making; time
geography; similarity measurement; group decision-making
1. Introduction
The capabilities of mobile navigation and communication devices are rapidly
improving. On the one hand, mobile routing services based on global positioning
system (GPS) support people during wayfinding. On the other hand, the advent of
mobile communication technologies such as general packet radio service (GPRS)
and universal mobile telecommunications system (UMTS) has enabled users of
mobile devices to access the internet anywhere at any time. The integration of both
within single devices has led to the development of a new generation of locationbased services (LBS). Knowing the position of the user, as well as the location of
available information sources allows for increasing the benefits of mobile services
dramatically.
Personalisation of services is a critical factor for improving the utility of LBS to
help people in making good decisions in their mobile everyday lives (Raper et al.
2007a). Recent research activities have focussed on various aspects of personalisation, that is, the customisation and adaptation of LBS to their users. This trend to
use highly specialised LBS has been intensified by people’s increased need to acquire
and use spatial and temporal information. In today’s world of vast mobility and
change we frequently face new situations in unfamiliar environments, such as finding
*Corresponding author. Email: [email protected]
ISSN 1748–9725 print/ISSN 1748–9733 online
ß 2009 Taylor & Francis
DOI: 10.1080/17489720903339668
http://www.informaworld.com
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
2
40
45
50
55
60
65
70
75
80
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
one’s way in an unfamiliar city. So far, the focus of research on personalised LBS has
been targeted at individual users, their situational and personal context and with
regard to mobile map applications (Schmidt et al. 1999, Raubal and Panov 2009).
However, location-based decisions are not only made by individuals but also by
groups of users (Golledge and Stimson 1997), which requires some kind of mediation
process among them.
The objective of this article is to investigate group decision-making in a mobile
context and provide a framework for the implementation of LBS, which support
groups of people in their everyday decision-making based on spatio-temporal
constraints and user preferences. According to the list of significant LBS research
issues presented by Raper et al. (2007b) we do not strive for presenting ad-hoc
solutions for improving LBS but to give an insight into the possibilities for future
services arising from the combination of several areas of research, such as time
geography, multi-criteria decision analysis (MCDA) and similarity measurement.
Location-based decision services (LBDSs) (Rinner and Raubal 2004) serve as the
foundation for this work. This idea and related research to personalised LBS are
outlined in the following section. In addition, a brief summary of time geography
and human decision-making is given. In Section 3 we introduce our use case for
location-based group decisions, that is, choosing a restaurant in the city of Münster.
Section 4 defines the optimal decision outcome of location-based group decisionmaking (LBGDM) according to the scenario. Based on this definition we define a
formal multi-criteria decision making (MCDM) model that allows for the calculation
and retrieval of such outcomes. In Section 5, the design and implementation of a
prototypical location-based group decision service realising the previously defined
MCDM model is described. This prototype is used for a simulation of a decisionmaking process according to the given scenario. The simulation and an evaluation of
the results are given in Section 6. The final section presents conclusions and
directions for future research.
2. Related work
The presented research focusses on two aspects of personalising LBS for groups
of users: optimising location-based information retrieval using methods of multicriteria evaluation and integrating spatio-temporal constraints using principles of
time geography. This section portrays the theoretical background for the LBGDM
model developed here. We start with a brief overview of human decision-making,
introduce the concept of a LBDS, and describe the principles of time geography.
2.1. Human decision-making
There is still insufficient knowledge about how mobile location-based decisionmaking is different from generic decision-making (Raper et al. 2007b). General
decision theory covers a wide range of models with different foci on describing how
decisions could or should be made, and on specifying decisions that are made
(Golledge and Stimson 1997, Amedeo et al. 2009). Human decision-making is not
strictly optimising in an economical and mathematical sense (Simon 1955), such as
proposed by the algorithms of classical decision-making theories, therefore
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
85
90
95
100
105
110
115
120
3
behavioural decision theory has been emphasised in the cognitive literature. In this
article, we implement a criterion-based judgment approach but several others exist.
A classification of decision-making models in terms of rationality is presented
in Todd and Gigerenzer (2000). They distinguish between unbounded rationality,
optimisation under constraints, satisfying, and their own approach of fast and frugal
heuristics.
In order to investigate whether principles of generic decision-making can be
transferred to mobile decision-making and find potential differences, researchers
have developed tools to study the interaction between environments, individuals and
mobile devices. Most case studies focus on pedestrian navigation in various settings,
such as urban environments (Li and Longley 2006). Raubal et al. (2004) proposed
a user-centred theory of LBS, which specifically focusses on individuals’ mobile
decision-making. It integrates spatial, temporal, social (using affordances) and
cognitive (using decision-making theory) aspects of LBS.
2.2. Location-based decision services
LBDSs are tools for supporting everyday decision-making in a mobile context. These
services are based on the integration of MCDA and can therefore provide analytic
evaluations of the attractiveness of alternative destinations and choices being offered
(Rinner 2008). MCDA methods had been introduced to geographic information
systems in the 1990s for applications, such as site suitability analysis (Malczewski
1999). Rinner and Raubal (2004) designed a service called Hotel Finder by
integrating the ordered weighted averaging (OWA) decision rule (Yager 1988).
This software features multi-criteria decision support for the task of finding suitable
hotels in an unfamiliar environment depending on the user’s location and
preferences. The usefulness of the Hotel Finder was demonstrated through a
comprehensive user test showing that MCDM-based decision support can be
employed for optimising location-based decision processes (Bäumer et al. 2007). The
results confirmed that applying the multi-criteria decision strategy enhances people’s
decision support in unfamiliar environments. Until now, LBDS were restricted to
single users and did not support groups of users in making location-based decisions.
Research in the field of spatial decision support systems (SDSS) has shown that
multi-criteria evaluation methods can also be applied to supporting group decision
processes that depend on spatial information (Jankowski 1997). Such systems have
been referred to as collaborative SDSS (Jankowski et al. 2006).
2.3. Time geography
An important aspect of personalised LBS is the integration of the users’ spatiotemporal constraints. Recent work has shown that principles of time geography
(Hägerstrand 1970) can be utilised to model spatio-temporal constraints in LBS
(Raubal et al. 2004). Time geography defines the space-time mechanics of people
and their environment by considering different constraints for people’s ability to be
present at a particular location in time – the capability, coupling and authority
constraints (Hägerstrand 1970). The possibility of being present at a specific location
and time is determined by people’s ability to trade time for space, supported by
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
4
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Figure 1. STP including PPS and PPA, according to Miller (1991b).
125
130
135
140
145
transportation and communication services. Space–time paths depict the movement
of individuals in space over time. Such paths are available at various spatial (e.g.
house, city, country) and temporal granularities (e.g. decade, year, day) and can be
represented through different dimensions. All space–time paths must lie within
space–time prisms (STPs) (Figure 1). These are geometrical constructs of two
intersecting cones and their boundaries limit the possible locations a path can take
based on people’s abilities to trade time for space. The inside of a STP is usually
referred to as potential path space (PPS), the projection of STPs onto geographic
space is called potential path area (PPA).
The graph-related adaptation of the STPs described by Miller (1991a) can be
applied to compute network time prisms (NTPs) for mobile agents in an urban
network. This was implemented (Raubal et al. 2007), realising a filtering technique
for selecting feasible host–client combinations in a shared-ride trip planning scenario
based on the intersection of potential path trees (PPT), which are the network
equivalents to PPAs.
3. Use case
The possibility for location-based group decisions occurs, if there is a group of users
whose STPs overlap for a specific time interval. To address such decision processes
we introduce the notion of LBGDM. LBGDM can deal with a variety of different
location-based tasks.1 The core question is thereby the same: ‘What is the best place
to perform a specific task?’
LBGDM addresses all components of geographic information (Goodchild
et al. 1999): temporal, spatial and thematic components. The selection of places for
location-based tasks greatly depends on the local knowledge of the decision maker.
This work focusses on cases of ad-hoc LBGDM in which the decision makers have
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
5
Table 1. User profiles for the task ‘going out for dinner’.
150
155
160
165
170
175
180
185
Group member
Priority
Choice
Location
Time limit (destination)
Alice
Bob
Charlie
Price
–
Cuisine
Chinese
Mexican
Mexican
Harbour
University castle
Hotel
–
8:30 pm (train station)
8:15 pm (back to hotel)
little knowledge of their environment and there are only limited resources for a
detailed investigation of the decision alternatives, for example, using a locationaware search engine, such as GoogleTM. The presented idea of LBGDM is based
on the concept of friend-finder services, which visualise the location of contacts on a
mobile map.2 These services enable groups of users to spontaneously meet to
perform common tasks, such as having dinner. The selection of a meeting point, as
well as a place for performing the task, for example, a restaurant, requires a group
decision process. In the following, we introduce the scenario for LBGDM to be used
for this work.
Alice is visiting the city of Münster (Germany) to attend a business meeting in the
area of the harbour. She has a free time slot in the evening and queries her friendfinder service to check whether any of her contacts are located in Münster. The
friend-finder service shows her the position of two of her contacts: Bob and Charlie.
Bob is currently attending a conference in the University castle and Charlie is in his
hotel room. Alice sends a short message to both and invites them to go out for
dinner. She has little knowledge of the city and its places. Therefore, she suggests
meeting at the train station and then discussing where to go. Each of the three
persons has different constraints and preferences. These can be summarised as user
profiles (Table 1). Alice likes Chinese food, but for her it is most important not to
pay too much for dinner. Bob and Charlie like Mexican food. While Bob has no
particular priority, it is very important for Charlie to have Mexican food.
They meet at the train station at 7:00 pm. Now they have to find out which
restaurants are nearby and meet their requirements. They start by querying the
search engines on their mobile web browsers and then discuss the results that are
retrieved. Alice has no more appointments this evening, but Bob has to catch a train
at 8:30 pm, and Charlie has to be back at the hotel in order to attend a conference
call at 8:15 pm. Deciding where to go takes longer than expected, by this time there is
only 1 h left until Charlie has to start back to his hotel. They decide to go to a nearby
rather expensive Italian restaurant to save time for dinner. What they missed was
that there are several nice restaurants close to Charlie’s hotel, which they could have
easily reached if they had met there directly and not at the train station. Decision
processes as described above can be time consuming and do not guarantee satisfying
outcomes. In addition, such processes are always influenced by the discussion
behaviour of the group. Each of the available options, for example, the restaurants in
the case described above, vary in suitability with respect to the overall set of
constraints and preferences. The larger the gap between individual interests, the
harder it is to find an overall satisfying solution. We argue here that such decision
processes can be supported by methods of MCDM.
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
6
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
4. A formal model of LBGDM
190
Before defining a MCDM model for any kind of decision support the objectives of
the decision process must be identified. The objective of the location-based decision
described in the previous section is to find the optimal place to have dinner. We use
the notion of optimal decision outcome to address this objective of LBGDM.
4.1. Optimal decision outcome
195
200
205
210
215
220
225
The optimal outcome of location-based decisions as illustrated in this work is defined
as the place that (1) does not violate any individual constraint, (2) is the best match
of the individual preferences and (3) is not a disadvantage for any group member.
The definition of the optimal decision outcome places requirements upon the model
to be developed:
(1) It must allow decision makers to formulate constraints, in particular, spatiotemporal constraints. Within MCDM constraints are sometimes considered
hard selection criterion (Rinner and Raubal 2004), in common language they
are often called K.O. criteria.
(2) It must provide quantitative methods for analysing decision alternatives with
respect to user preferences.
(3) It requires mediation functionality to ensure that no group member is
disadvantaged. If an alternative performs well for a majority of group
members but badly for a minority, it has to be considered less optimal than
an alternative that ensures that each group member is at least satisfied, even
if the performance is worse for the majority.
4.2. Spatio-temporal screening
In an area, such as the city of Münster the multitude of restaurants and bars provide
many alternatives for decision processes, such as illustrated earlier. The more
identifiable alternatives there are, the more complex the decision problem becomes.
The first step of the proposed analysis aims at screening those alternatives that are
not feasible with respect to spatio-temporal constraints. As mentioned in Section 2.3,
Hägerstrand (1970) identified three categories of spatio-temporal constraints:
capability, coupling and authority constraints. Capability constraints limit the activity
possibilities of an individual due to available resources. Depending on the mode of
transportation an individual can have different capability constraints. In our
scenario, capability constraints are defined by the walking speed of the individual
group members.3 Coupling constraints limit the activity possibilities of an individual
by requiring the concurrent presence with other individuals at one place for a specific
time interval. In LBGDM, coupling constraints are set by the personal schedule, for
example, meetings or departure times, and the stops that allocate a task for a given
time in a specific place.4 Authority constraints limit the accessibility of places for
specific time intervals. Restaurants, for example, impose authority constraints on
their guests through their opening hours.
To account for spatio-temporal constraints we adopt the network-based
algorithmic solution presented in Raubal et al. (2007). It optimises the selection of
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
230
235
240
245
250
7
shared rides in a transportation network depending on the clients and hosts’ STPs.
The underlying problems of determining the set of feasible hosts in shared-ride trip
planning and determining the set of feasible alternatives in LBGDM are similar. In
both cases an assessment whether a particular location, either a host’s position or a
place alternative, is reachable within a given time must be made. For LBGDM such
a filtering mechanism should yield those nodes of a street network for the area of
interest at which feasible places are located. The particular method involves the
computation of PPT conceptualised as the network equivalents of the se (start,
earliest)-cone and the dl (destination, latest)-cone (Winter and Raubal 2006). The
overlap of both trees results in a NTP (Wu and Miller 2001), the network equivalent
of the general STP. To adopt this approach for LBGDM the relevant se- and
dl-cones must be identified. If a group decides to perform a location-based task, its
members typically agree on a start time, for example, dinner at 7:00 pm. This article
focusses on such cases, in which the individual group members are not co-present at
one location, but spread across an urban area. For LBGDM the se-cone is defined as
the part of the space–time continuum that is reachable from the start point in space
(and time) and the time difference from this start point and the beginning of the first
task (Figure 2a). If the task starts at 7:00 pm and a group member can start travelling
at 6:30 pm, the se-cone for this member is the sub-tree of the street network that is
reachable within 30 min from the start node. The dl-cone for LBGDM has to be
considered if a group member has a particular destination she must reach by a given
time after meeting with the group (Figure 2b). She might, for example, need to catch
a train at 8:30 pm. Destinations in space and time are the basis for creating the dlcone. The size of the dl-cone is determined by the time that goes by between the end
Figure 2. Spatio-temporal filtering: (a) se-cone, (b) dl-cone and (c) NTP.
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
8
255
260
265
270
275
280
M. Espeter and M. Raubal
time of the task and the time limit. If dinner ends at 8:00 pm there are 30 min left to
get from the restaurant to the train station. With a sequence of tasks, the end of the
final task must be considered for creating the dl-cone. The overlap of both cones
leads to a NTP to be used for filtering the feasible places with respect to one group
member’s individual constraints. Overlapping the individual NTPs yields the feasible
places with respect to the group’s overall constraints (Figure 2c).
4.3. Stop assessment
After the set of feasible places has been determined the viability of each place with
respect to the users’ preferences and non-spatio-temporal constraints must be
assessed. Besides dealing with constraints, such as price limits, this process also
requires scores for the relevant decision criteria, based on compensatory MCDM
techniques. In the case of the task ‘going out for dinner’ the criteria may be prices,
place ratings and cuisine. To fulfil the second definition of the optimal decision
outcome for LBGDM it must be assessed how well each alternative performs with
respect to the user preferences. This requires the use of a decision rule. Decision rules
are compensatory procedures for ordering decision alternatives according to their
performance with respect to the objective (Malczewski 1999). In this case the
objective is characterised by the definition of the optimal decision outcome.
Compensatory MCDM techniques require numerical and standardised decision
variables (Jankowski 1995). Depending on the scale of the criteria the application of
such techniques requires transformations. Ratio scales, the scale of choice for
modelling price, generally are not standardised. The standardisation of ratio
variables can be achieved by applying linear scale transformation. For this purpose
we use the score range procedure, which ensures that the whole range between 0 and 1
is used. In the case of benefit criteria, such as place ratings, raw values are
standardised by dividing the difference between the raw value and the minimum
value by the range (F 1). Since place ratings are modelled using ordinal scales, the
raw values for applying formula 1 refer to the position (index) of a particular value
within the given range. In the case of cost criteria, such as price limits this formula
needs to be modified slightly (F 2).
ðF1Þ vnorm ¼
285
290
295
[1–23]
[PREPRINTER stage]
vraw vmin
vmax vmin
ðF2Þ vnorm ¼
vmax vraw
vmax vmin
In both cases a value of 0 is given to the worst possible score, whereas a value of 1
is given to the best possible score. To account for the spatial distribution of
alternatives, minimum and maximum values are taken from those alternatives within
the NTP that is determined by the group’s spatio-temporal constraints. The
assessment of each variable is therefore relative to the other variables of the same
type provided by the surrounding alternatives.
Integrating attributes on nominal scales, for example, cuisines into a compensatory MCDM approach is not straightforward. In general, nominal scales only
provide information on whether two values are identical or not. We therefore
propose the use of similarity measurement methods for integrating nominal data into
a compensatory MCDM process. If people discuss where to go for dinner they often
decide about the cuisine that is offered by a place, for example, they like Chinese or
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
300
305
310
315
320
9
Mexican food. According to the first law of geography (Tobler 1970) ‘everything is
related to everything else, but near things are more related than distant things’. This
can also be assumed for cuisines on a world-region scale. The Japanese and the
Chinese cuisines are more related than the Japanese and the Mexican cuisines. Such
relatedness can be derived by the common sense classification of world regions,5
which can be used to classify restaurants according to the region whose cuisine they
provide. The open directory project6 uses a similar classification for structuring
recipes on their web portal. To quantify the relatedness of two cuisines, we utilise the
hierarchical structure provided by those classifications. Hierarchies can be modelled
as concept graphs consisting of nodes that are connected via is-a links (sub-concept
relations). The semantic distance between concepts in a hierarchy can be employed
to assess the semantic similarity of two concepts (Rada et al. 1989). Figure 3 shows
a graphical view of the cuisine hierarchy used in this model.
This article presents an approach to measure the similarity between a cuisine
objective cobj and a cuisine alternative calt based on the distance measure dist(cobj,
calt) between the nodes that allocate the cuisines in the hierarchy (F 3).
8
if cobj subsumes calt
>
<1
if lubðcobj , calt Þ ¼ top :
ðF3Þ simðcobj , calt Þ ¼ 0
>
:
1=distðcobj , calt Þ þ 1 else:
If the cuisine alternative is subsumed by the cuisine objective (e.g. Chinese cuisine is
subsumed by Asian) the similarity value is 1. Hence, the proposed similarity measure
is asymmetric in the way that a super concept is considered equal to its sub concepts,
which is inverse to the way asymmetry is usually considered (Tversky 1977). The
proposed understanding of asymmetry is reasonable, since people may want to select
different levels of the hierarchy. If a person wants an Asian dinner she does not
bother whether the restaurant provides Chinese or Indian food. If the most specific
common super concept or least upper bound (lub) (Rodrı́guez and Egenhofer 2003)
of cobj and calt is the top-concept, the similarity score is 0 (e.g. Chinese to German).
Similarity is a decaying function of distance, therefore the inverse distance is taken as
input for the decision analysis. This leads to a value range between 0 and 1, which is
Figure 3. Taxonomy for measuring similarity between cuisines (subtree).
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
10
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Table 2. Selected similarity scores.
325
330
335
Objective
Alternative
Similarity score
Chinese
Chinese
Chinese
Chinese
Chinese
Asian
Chinese
East-Asian
Japanese
Indian
German
Chinese
1
0.67
0.5
0.25
0
1
an implicit standardisation. Additionally, the distance is always incremented by 1 to
ensure that a distance of 0 leads to a similarity value of 1. The resulting similarity
values are strongly influenced by the weights that are assigned to the links of the
hierarchy. We choose here a value of 0.5 to achieve reasonable results. Table 2 shows
selected similarity scores for measuring similarity between cuisines.
After the relevant criteria scores have been computed, the decision alternatives
can be assessed by considering the user preferences. We use the point allocation
weighting method, which allows each group member to allocate 100 points to the set
of decision criteria, that is, price, rating and cuisine. The scores combined with the
individual weights are aggregated via the simple additive weighting (SAW) technique,
an additive decision rule based on weighted arithmetic mean computation. SAW is
sometimes called scoring or weighted linear average and is one of the most frequently
used techniques for spatial multiple attribute decision-making. SAW scores each
decision alternative A with respect to a user j by summing up the product of each
criterion score Ci and the associated weight wi for m relevant decision criteria (F4).
ðF4Þ Aj ¼
m
X
wi Ci :
i¼1
340
The input for SAW is typically specified via decision tables that store the criterion
scores for each alternative. Applying SAW results in a ranked list of place
alternatives for each user.
4.4. Mediation
345
350
To assess the performance of a stop for the whole group the individual stop scores
have to be aggregated. One could simply use arithmetic means, as done by the SAW
method. However, the application of arithmetic means for assessing group stops
does not guarantee optimal decision outcomes. Arithmetic means can lead to a
violation of the third requirement (Section 4.1), namely that no group member is
disadvantaged. If one place provides high scores for two of the group members (e.g.
0.8 and 0.9) but a low score for a single one (e.g. 0.2), the arithmetic mean gives the
relatively high value of 0.63. The result implies that this alternative performs well
with respect to the group’s interests. It conceals that the third group member may not
be satisfied by this selection. If there is another alternative that achieves scores of 0.6
for each user, the arithmetic mean will be 0.6, which is lower than the score assigned
to the first alternative. According to the definition of the optimal decision outcome,
the latter alternative should be preferred. In other words, a mediation process
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
11
Figure 4. Overview of LBGDM model.
355
between user preferences is required. Therefore, we suggest another aggregation
technique for stop assessment of groups: the computation of harmonic means (Bullen
2003). The score of one stop S is the harmonic mean of all alternative scores A for
each of n users (F5).
n
ðF5Þ S ¼ Pn 1 :
j¼1 Aj
360
365
370
Using harmonic means leads to a larger impact of low scores on the result. In the
example mentioned above the harmonic mean of the first alternative is 0.48, whereas
the score of the second alternative is still 0.6.
After the mediation process has been performed the alternatives are ranked so
that the best alternative fulfils the requirements of the optimal decision outcome.
Taking it a step further, location-based group decision support service (LBGDS)
could also be designed to account for multiple tasks, for example, start with dinner
and continue with going to the movies or having some drinks. Single user decision
support for such task combinations was realised in the project utopian.7 To keep the
possibility for supporting multiple tasks open, we refer to the result of the decision
process as a tour. A tour can comprise one or more stops that allocate particular
tasks to given places, as well as the paths the users have to follow to get to the stops
and their potential destinations.
Figure 4 summarises the methods presented in this section for the case of two
users and shows how they can be combined into a model that supports the retrieval
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
12
375
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
of the optimal decision outcome in a location-based group decision process as
defined in Section 4.1.
5. Prototype
380
385
390
395
400
405
410
415
In the following, the presented model will be realised and evaluated in a prototypical
LBGDS called mediatrix.8
The classic approach for developing LBS is to deploy the whole application on
the mobile device without requiring network access (fat-client). An example for a fatclient is the first version of the Hotel Finder (Raubal and Rinner 2004). This solution
is suitable for simple problems involving a small data set, but not for LBS used to
answer complex spatio-temporal queries. Recent developments of mobile devices,
in particular with respect to their communication capabilities, have offered the
possibility to develop LBS that also involve server components to be accessed over a
network. The latest version of the Hotel Finder (Bäumer et al. 2007) is an example of
such distributed LBS. These systems still require the installation of service-specific
client software on the mobile device. In the context of mobile applications this
requirement is especially critical. The market for personal digital assistants (PDAs)
and Smartphones has become highly dynamic and is competitive with contenders
offering ambitious platforms, such as the AppleTM iPhone, GoogleTM Android,
WindowsTM Mobile and SymbianTM (S60). Software providers have to adjust their
applications to each platform they want to support, which can be a time consuming
task. A novel approach for designing LBS is to utilise web browsers as client
components. This has recently gained importance in the Internet community and
benefited from the evolution of web browser technology. One of the most important
results was the creation of the AJAX – pattern (asynchronous javascript and XML)
(Garrett 2005). AJAX describes the development of web applications unifying the
benefits of classical web and desktop applications. They provide sophisticated
functionality and a dynamic user interface with only a web-browser on the client. At
the core of AJAX is the asynchronous XMLHTTPRequest that allows reloading
only specific parts of a web page instead of the whole page as done by classic
web applications. Prominent examples for current AJAX-applications are Google
MapsTM and FlickrTM.
5.1. Client
For the implementation of mediatrix we followed the AJAX-approach. The client is
realised using hypertext markup language (HTML), cascading style sheets (CSS) and
JavaScript. The server component is based on JavaTM Servlet Technology and the
Apache Wicket web application framework. The user interface is optimised for a
resolution of 240*320 pixels, which is the de facto standard for displays of current
mobile communication devices. Apache Wicket has been extended by some AJAXenabled components, such as drop-down choices and links that allow reacting to user
input by dynamically reloading other page components. This allows the development
of a dynamic web front-end with the look-and-feel of a fat-client. To access mediatrix
the user must visit the homepage and log in. After the log-in, the user gets to the
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
420
425
430
435
440
445
13
index page, where she can create a tour request, join an existing tour request, edit her
profile or view computed tour proposals.
One of the advantages of using current AJAX applications is the possibility
to combine various online data sources to one hybrid application, called mash-ups
(Scharl 2007). Google MapsTM has proven useful for mash-ups that require the
visualisation of geographic information. For mediatrix the localisation of the users,
their destinations and the visualisation of the computed tour have been realised this
way. The client component has been tested on a PDA equipped with Mozilla
Minimo which is a mobile browser that supports AJAX-based web applications.
5.2. Data
LBGDS require several types of data: network data, user data, place data and
request data. The network data are provided by TeleAtlasÕ and stored in a relational
database (RDB) consisting of node and edge tables. User profiles and their inputs
are mapped onto a RDB schema and also stored in a RDB. The RDB further stores
the place data, which consists of spatio-temporal and thematic data. The thematic
data include information on different scales. Ratio and ordinal data (e.g. prices and
ratings) are stored using standard structured query language (SQL) data types. These
are not sufficient to store the nominal data required for the decision analysis. Section
4.3 demonstrated how nominal data can be standardised using similarity
measurement. This requires data that is structured hierarchically. For this purpose,
the cuisine data is stored in the RDB, but also linked to a cuisine hierarchy that is
defined using resource description framework-schema (RDFS). RDFS extends
resource description framework (RDF) by a standard vocabulary and some core
semantics. In particular, it provides the possibility to model hierarchies that are
required for similarity measurement.
The current version of mediatrix supports group decision processes for up to
three users with respect to the task of going out for dinner. The prototype accounts
for spatio-temporal and price constraints, as well as for the opening hours of places.
The integration of further data for planning tasks, such as going to the movies or
having a drink, and the possibility to combine tasks is left for future work. The user
interface has been designed to select more than one task.
6. Evaluation
450
455
This section discusses the model and prototype that have been presented in the
previous sections. It includes a simulation of mediatrix with respect to the scenario
described in Section 3. The results of this simulation are evaluated and used to assess
whether the proposed model is able to achieve optimal decision outcomes.
6.1. Simulation
The scenario underlying this work presents an example for a group decision problem
in which a group of three people with limited spatio-temporal resources tries to find
a restaurant in the city of Münster. The group is characterised by three user profiles
(Table 2). These profiles are now taken as user inputs for running a decision process
supported by mediatrix. According to the scenario, Alice starts the group task.
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
14
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Figure 5. Create tour.
460
465
470
475
480
She wants to invite two of her friends, Bob and Charlie, to have dinner at 7:00 pm.
Therefore, she logs on to mediatrix and creates a tour request with the required
parameters start time, invited users and task type (Figure 5).
After the request is registered by the service, it gets published to the invited users.
At the current state of the prototype the users have to log in to check for tours they
are invited to. For future versions it would be reasonable to integrate a notification
service, for example, a short message service (SMS), which directly contacts users
after they have been invited to a tour. To join a tour request the invited users can
select the tour name and view its details. They can either join the tour or decline the
invitation. Figure 6 shows how Bob reacts to the invitation by Alice.
After joining a tour request the user must submit her personal input to the
service, that is, the relevant constraints and preferences (Figure 7). Figure 8 shows
the Google MapsTM mash-up that is used for setting the start and destination. The
input for a tour request is enriched by the personal user profiles that store static and
request-independent data, such as weights (Figure 9). The profiles are loaded from
the RDB. To perform the decision analysis, each user must submit her constraints
and preferences.9 Table 3 summarises the concrete user inputs for the tour request
created by Alice.
After the group’s input is complete it is analysed with respect to the area of
interest. According to Section 4.2, the analysis starts with spatio-temporal screening
resulting in a group NTP that contains the set of feasible places. Table 4 lists the
feasible places for the given request and the place data that is important for
the decision analysis. To assess the price level of restaurants, the minimum and the
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
Figure 6. Join tour.
Figure 7. Set input.
15
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
16
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Figure 8. Set spatial input.
485
490
495
maximum price for a single dish are taken to calculate an average price. This is a
simple, but plausible approximation of the price level which could be extended by
taking all offers provided by a place into account.
Next, the relevant decision criteria are scored including normalisation and
similarity measurement. In this case the similarity measurement affects the cuisines
provided by the feasible places and the choices made by each group member. The
raw values have to be normalised and then aggregated with respect to the user
weights using the SAW method. The results are shown in Table 5.
Finally, the place scores are aggregated to stop scores including the input of
the whole group. The decision strategy proposed here utilises harmonic mean
computations for this step (Section 4.4). Table 6 shows the stop scores computed
with harmonic mean compared to the geometric mean computation. The Enchilada
achieves the highest score using arithmetic mean, whereas the Ipanema scores best
with respect to the harmonic mean. China Corner and Altes Gasthaus Leve achieve
a score of 0 using harmonic mean. The resulting tour is shown in Figure 10.
6.2. Results
The results of the simulation are now analysed in the context of the definition of the
optimal decision outcome. The optimal decision outcome of location-based group
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
17
Journal of Location Based Services
Figure 9. Edit profile.
Table 3. User profiles for the task ‘having dinner’.
User
Alice
Bob
Charlie
Choice
Price weight
Rating weight
Choice weight
Chinese
Mexican
Mexican
80
30
0
10
30
0
10
40
100
Table 4. Feasible places and their attributes.
Name
Cuisine
Altes Gasthaus LeveGerman
China Corner
Chinese
Enchilada
Mexican
Ipanema
Latin American
RatingMin price (E)Max price (E)
4.38
4.75
3.75
2.5a
8
3.50
8.90
4.5
22
8
15.30
12
Note: aNo rating available on www.restaurant-kritik.de (2.5 is the
default rating).
500
decisions is defined in Section 4.1 as the outcome that does not violate any individual
constraints and matches the preferences best without causing disadvantage to any
group member. The first requirement, no violation of user constraints, is achieved by
each feasible place. In order to confirm the second requirement the place scores have
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
18
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Table 5. Aggregated place scores for each user.
Place
Alice
Bob
Charlie
Altes Gasthaus Leve
China Corner
Enchilada
Ipanema
0.08
1
0.33
0.63
0.27
0.59
0.72
0.64
0
0
1
0.67
Table 6. Aggregated stop scores.
Place
Ipanema
Enchilada
China Corner
Altes Gasthaus Leve
505
510
515
520
525
530
Score (arithmetic)
Score (harmonic)
0.65
0.68
0.52
0.11
0.65
0.55
0
0
to be considered. Due to its Mexican cuisine the Enchilada achieves high scores for
Bob and Charlie but because of the high price level a low score for Alice. Taking the
arithmetic mean for aggregating the user-specific place scores, the Enchilada achieves
the highest score and accordingly fulfils the second requirement. However, Alice
would not be happy with this solution and therefore the Enchilada violates the third
requirement. Using harmonic mean computation yields the Ipanema as the best
result. Since the Ipanema is less expensive, Alice can be quite satisfied with having
dinner there. For Charlie it is very important to have Mexican food. The Ipanema
provides Latin American cuisine, which is not the same but similar. For Bob both
places provide rather satisfying results. Taking everything into account it can be
concluded that the Ipanema is the best compromise that could be achieved for the
given decision problem and is therefore considered as the optimal decision outcome.
It follows that the proposed MCDM model allows for finding optimal decision
outcomes for location-based group decision processes.
6.3. Discussion
Several challenges remain for realising LBGDS according to the model presented.
First, one must investigate how closely human decision-making behaviour in
scenarios as described in this work can be approximated by the proposed model. This
requires a user-centred focus to investigate particular needs of the user, how they
interact with other users in a group, and how they make judgments and trade these
off. Such research has a long tradition and has been applied within the area of LBS
with a focus on individual users and user groups (Wealands et al. 2007, Raubal 2009)
for different scenarios, such as mobile guides (Baus et al. 2005). In particular and
regarding the work presented here, this affects the definition of the optimal decision
outcome, which has not yet been verified with respect to human judgements. The
identified decision criteria require further investigation. According to Malczewski
(1999), a set of MCDM criteria has to be comprehensive, measurable, complete,
operational, decomposable, non-redundant and minimal. Verification that these
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
19
Figure 10. Final result for dinner tour.
535
540
545
characteristics are provided by the set of MCDM criteria utilised in this work has
not yet been achieved. We must also assume that some additional criteria, such as
ambiance or place size, may play important roles for location-based decisions.
The dynamic aspects of LBGDS need further investigation, as well. Currently,
constant travel velocity is assumed and transportation networks, such as for buses
have not been regarded yet. Positioning via self-localisation in the current version of
mediatrix should be replaced by automated positioning techniques and geocoding
services. Finally, it has to be stated that our approach does not strive for guaranteed
perfect results, that is, very high scores for each user in general. The outcome of
location-based decision processes can rarely be perfect, because user interests vary
and the set of alternatives is strongly restricted by the spatio-temporal context.
Hence, the underlying notion of optimal should rather be understood in terms of best
possible. One of the core advantages of mediatrix is that it only requires asynchronous
telepresence of group members (Harvey and Macnab 2000). There is no need for
direct interaction as in a standard discussion process since the service maintains
requests that have been created until all users have submitted their input.
7. Conclusions and future work
This article presented a formal model and prototypical implementation for LBGDM.
It provides a foundation for the future development of personalised LBS that
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
20
550
555
560
565
570
575
580
585
590
595
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
support groups of users in their everyday planning of shared activities. Such
applications include any location search for common activities, collaborative
emergency response operations, mobile guides and the broad range of social
network functions. We outlined examples and basic principles of personalisation in
LBS, and presented a use case for LBGDM dealing with a group of people searching
for a restaurant in the city of Münster. A definition of the optimal decision outcome
for LBGDM was given. The formal MCDM model aimed at obtaining such
outcomes depending on personal constraints and preferences. Principles of time
geography were used to develop a screening method that accounts for spatiotemporal constraints. Similarity measurement methods were employed to integrate
nominal decision criteria into decision rules. We proposed using harmonic means
to avoid individual group members being disadvantaged. By simulating a decision
process using the prototypical LBGDS mediatrix we demonstrated that the proposed
model can be employed to achieve optimal decision outcomes for location-based
group decisions.
There are several areas for future research with regard to improving the model
presented in this work: spatio-temporal attributes have been considered as noncompensatory so far. Allowing compensatory spatio-temporal attributes would
enable users to express preferences with respect to spatial-temporal criteria, for
example, some users may prefer short travel distances. The proposed MCDM model
utilises a series of simple techniques. More sophisticated methods are available, for
example, the OWA method. OWA allows for selecting between different personal
decision strategies and has already been implemented in LBDS for single users
(Rinner and Raubal 2004). Using OWA the user can select optimistic, pessimistic
and moderate decision strategies. An optimistic strategy puts higher weights on
criteria with high scores, whereas a pessimistic strategy puts higher weights on
criteria with low scores. Another MCDM technique that could be utilised for
LBGDM is sensitivity analysis (SA) (Malczewski 1999, Ligmann-Zielinska and
Jankowski 2008). Using SA in a LBGDS could provide helpful feedback for decision
makers by showing how sensitive the selected criteria are with respect to changes in
user input. It could be combined with a voting mechanism offering several wellperforming tour alternatives. For both MCDM techniques, OWA and SA, the
high cognitive workload demanded from the user seems to be a major obstacle for
employing them for location-based group decision support. As described in Section
6.3, a user-centred focus will bring more insight to these questions.
The similarity measure proposed here has not been evaluated with respect to its
correspondence to human similarity judgements. Considering the semantic distance
method employed for measuring cuisine similarity it may be reasonable to include
further similarity measures, for example, to exploit information contents as proposed
by Resnik (1995). Resnik suggests using probabilities of encountering an instance of
a concept for measuring similarity in taxonomies. He defines information content as
the negative logarithm of this probability. The similarity between two concepts C1
and C2 is measured using the information content of the most specific class, or least
upper bound, subsuming C1 and C2. For the top-class of the taxonomy the
probability of encountering an instance is 1, which leads to a similarity measure of 0.
This is the result of comparing two concepts located in different main-branches of
the taxonomy, for example, Chinese to German cuisine. The probability computation depends on empirical, instance-based information. In the case of measuring
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
600
605
610
21
cuisine similarity, this translates to restaurant data. If there were many East-Asian
restaurants (40% probability) and few African restaurants (5% probability) there
would be a higher similarity between African restaurants, for example, sim
(Tunesian, Egyptian), than between East-Asian restaurants, for example, sim
(Japanese, Chinese), which seems plausible. A combination of the approach used
in this work and the information content may improve the decision analysis.
Our major objective was to define a formal, computational model for enabling
location-based decision support for user groups. This strongly influenced the
development of the prototype, which was developed to demonstrate and analyse this
model. As a next step, an evaluation of how such a service could be embedded in the
service-oriented architecture provided by the Open-LS specifications10 is necessary.
Mediatrix has been designed and implemented as a stand-alone web application
in the Web 2.0 style and does not provide interfaces to connect to other services.
Aligning mediatrix to the Open-LS specification would allow accessing other
services, such as geocoding and navigation services, which are appropriate extensions
for a LBGDS.
Acknowledgements
615
The comments from the anonymous reviewers and the editors provided useful suggestions
to improve the content of the article. Funding by the German Academic Exchange Service
(DAAD) for the first author is gratefully acknowledged.
Notes
620
625
630
1. We define location-based tasks as activities that are bound by space and time. For
example, eating is an activity, whereas having dinner is a task.
2. For example, http://www.google.com/latitude, www.qiro.de, www.kakiloc.com
3. For reasons of simplicity a constant travel velocity (5 km/h) is assumed for each group
member.
4. We assume a standard duration of 1 h for the task of having dinner.
5. http://unstats.un.org/unsd/methods/m49/m49.htm
6. http://dmoz.org/Home/Cooking/World_Cuisines
7. http://www.youtube.com/watch?v¼m0zD_n3zEYk
8. mediatrix (Latin): female troubleshooter.
9. This simulation focusses on quantitative methods. Thus, no additional constraints, such
as price limits are set.
10. http://www.opengeospatial.org/standards/ols
References
635
Amedeo, D., Golledge, R., and Stimson, R., 2009. Person environment behavior research:
investigating activities and experiences in spaces and environments. New York: Guilford
Press.
Bäumer, B., Panov, I., and Raubal, M., 2007. Decision improvement through multi-criteria
strategies in mobile location-based services, In: A. Car, G. Griesebner and J. Strobl, eds.
Geospatial crossroads @ GI_Forum: proceedings of the first geoinformatics forum
Salzburg. Heidelberg: Wichmann, 17–25.
1
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
22
640
645
650
655
660
665
670
675
680
685
690
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
M. Espeter and M. Raubal
Baus, J., Cheverst, K., and Kray, C., 2005. A survey of map-based mobile guides, In: L. Meng,
A. Zipf and T. Reichenbacher, eds. Map-based mobile services. Berlin: Springer,
197–216.
Bullen, P., 2003. Handbook of means and their inequalities. Boston, MA: Kluwer.
Garrett, J., 2005. Ajax: a new approach to web applications. Adaptive Path. [online]. Available
from: http://www.adaptivepath.com/ideas/essays/archives/000385.php
Golledge, R. and Stimson, R., 1997. Spatial behavior: a geographic perspective. New York:
Guilford Press.
Goodchild, M., et al., 1999. Introduction to the Varenius project. International Journal of
Geographical Information Science, 13, 731–745.
Hägerstrand, T., 1970. What about people in regional science?. Papers of the Regional Science
Association, 24, 7–21.
Harvey, A. and Macnab, P., 2000. Who’s up? Global interpersonal accessibility, In: D. Janelle
and D. Hodge, eds. Information, place and cyberspace: issues in accessibility. Berlin:
Springer, 147–170.
Jankowski, P., 1995. Integrating geographical information systems and multiple criteria
decision-making methods. International Journal of Geographical Information Systems,
9, 251–273.
Jankowski, P., 1997. Spatial group choice: a SDSS tool for collaborative spatial decision
making. International Journal of Geographical Information Science, 11, 577–602.
Jankowski, P., et al., 2006. Design considerations and evaluation of a collaborative, spatiotemporal decision support system. Transactions in GIS, 10, 335–354.
Li, C. and Longley, P., 2006. A test environment for location-based services applications.
Transactions in GIS, 10, 43–61.
Ligmann-Zielinska, A. and Jankowski, P., 2008. A framework for sensitivity analysis in spatial
multiple criteria evaluation. In: T. Cova, H. Miller, K. Beard, A. Frank, &
M. Goodchild, eds. Geographic information science, 5th International conference,
GIScience 2008, September 2008, Park City, UT, USA. Berlin: Springer, 217–233.
Malczewski, J., 1999. GIS and multicriteria decision analysis. New York: John Wiley.
Miller, H., 1991a. Modeling accessibility using space-time prism concepts within geographical
information systems. International Journal of Geographical Information Systems, 5,
287–301.
Miller, H.J., 1991b. Modelling accessibility using space-time prism concepts within
geographical information systems. International Journal on Geographical Information
Science, 5, 287–301.
Rada, R., et al., 1989. Development and application of a metric on semantic nets. IEEE
Transactions on Systems, Man and Cybernetics, 19, 17–30.
Raper, J., et al., 2007a. Applications of location-based services: a selected review. Journal of
Location Based Services, 1, 89–111.
Raper, J., et al., 2007b. A critical evaluation of location based services and their potential.
Journal of Location Based Services, 1, 5–45.
Raubal, M., 2009. Cognitive engineering for geographic information science. Geography
Compass, 3, 1087–1104.
Raubal, M., et al., 2007. Time geography for ad-hoc shared-ride trip planning in mobile
geosensor networks. ISPRS Journal of Photogrammetry and Remote Sensing, 62,
366–381.
Raubal, M., Miller, H., and Bridwell, S., 2004. User-centred time geography for locationbased services. Geografiska Annaler B, 86, 245–265.
Raubal, M. and Panov, I., 2009. A formal model for mobile map adaptation, In: G. Gartner
and K. Rehrl, eds. Location based services and telecartography II – from sensor fusion
to context models. Selected papers from the 5th international symposium on LBS &
TeleCartography 2008, Salzburg, Austria. Berlin: Springer, 11–34.
2
XML Template (2009)
{TANDF}TLBS/TLBS_A_434144.3d
[7.10.2009–9:36pm]
(TLBS)
[1–23]
[PREPRINTER stage]
Journal of Location Based Services
695
700
705
710
715
720
725
23
Raubal, M. and Rinner, C., 2004. Multi-criteria decision analysis for location based
services. In: S.A. Brandt, ed. Geoinformatics 2004, 12th International conference on
geoinformatics, 7–9 June 2004, Gävle, Sweden. Gävle: University Press, 47–53.
Resnik, P., 1995. Using information content to evaluate semantic similarity in a taxonomy.
Proceedings of the 14th international joint conference on artificial intelligence. Montreal,
Canada, 448–453.
Rinner, C., 2008. Mobile maps and more – extending location-based services with multicriteria decision analysis, In: L. Meng, A. Zipfand and S. Winter, eds. Map-based mobile
services. Berlin: Springer, 335–352.
Rinner, C. and Raubal, M., 2004. Personalized multi-criteria decision strategies in locationbased decision support. Journal of Geographic Information Sciences, 10, 149–156.
Rodrı́guez, A. and Egenhofer, M., 2003. Determining semantic similarity among entity
Classes from different ontologies. IEEE Transactions on Knowledge and Data
Engineering, 15, 442–456.
Scharl, A., 2007. Towards the geospatial web: Media platforms for managing geotagged
knowledge repositories, In: A. Scharl and K. Tochtermann, eds. The geospatial web.
London: Springer, 3–14.
Schmidt, A., Beigl, M., and Gellersen, H., 1999. There is more to context than location.
Computers and Graphics, 23, 893–901.
Simon, H., 1955. A behavioral model of rational choice. Quarterly Journal of Economics, 69,
99–118.
Tobler, W., 1970. A computer movie simulating urban growth in the Detroit region. Economic
Geography, 46, 234–240.
Todd, P. and Gigerenzer, G., 2000. Précis of simple heuristics that make us smart. Behavioral
and Brain Sciences, 23, 727–780.
Tversky, A., 1977. Features of similarity. Psychological Review, 84, 327–352.
Wealands, K., et al., 2007. User assessment as input for useful geospatial representations
within mobile location-based services. Transactions in GIS, 11, 283–309.
Winter, S. and Raubal, M., 2006. Time geography for ad-hoc shared-ride trip planning.
In: K. Aberer, T. Hara and A. Joshi, eds. 7th International conference on mobile data
management (MDM’06). Nara, Japan: IEEE.
Wu, Y.-H. and Miller, H., 2001. Computational tools for measuring space-time accessibility
within dynamic flow transportation networks. Journal of Transportation and Statistics,
4, 1–14.
Yager, R., 1988. On ordered weighted averaging aggregation operators in multi-criteria
decision making. IEEE Transactions on Systems, Man and Cybernetics, 18, 183–190.
3
4